Samuel Eckmann
@samueleckmann.bsky.social
Computational neuroscientist, currently in Cambridge, UK (CBL) as a Newton International Fellow. Interested in neural circuits, E-I balance, and biological learning → http://samueleckmann.github.io
"Can an AI model predict perfectly and still have a terrible world model? [...] A transformer trained on 10M solar systems nails planetary orbits. But it botches gravitational laws"
Twitter thread: x.com/keyonV/statu...
Paper: arxiv.org/abs/2507.06952
Twitter thread: x.com/keyonV/statu...
Paper: arxiv.org/abs/2507.06952
July 13, 2025 at 1:20 PM
"Can an AI model predict perfectly and still have a terrible world model? [...] A transformer trained on 10M solar systems nails planetary orbits. But it botches gravitational laws"
Twitter thread: x.com/keyonV/statu...
Paper: arxiv.org/abs/2507.06952
Twitter thread: x.com/keyonV/statu...
Paper: arxiv.org/abs/2507.06952
Flow fields never get old
April 27, 2025 at 9:20 PM
Flow fields never get old